I don't know what your code is doing exactly. For comparison, here's my utf8 decoder (for a single codepoint):

    static UnicodeCodepoint utf8_decode(u8 const bytes[static 4], u8 *out_num_consumed) {
        u8 const flipped = ~bytes[0];
        if (flipped == 0) {
            // Because __builtin_clz is UB for value 0.
            // When his happens, the UTF-8 is malformed.
            *out_num_consumed = 1;
            return 0;
        }
        
        u8 const num_ones = __builtin_clz(flipped) & 0x07;
        u8 const num_bytes_total = num_ones > 1 ? num_ones : 1;
        u8 const main_byte_shift = num_ones + 1;
        UnicodeCodepoint value = bytes[0] & (0xFF >> main_byte_shift);
        
        for (u8 i = 1; i < num_bytes_total; ++i) {
            if (bytes[i] >> 6 != 2) {
                // Not a valid continuation byte.
                *out_num_consumed = i;
                return 0;
            }
            
            value = (value << 6) | (bytes[i] & 0x3F);
        }

        *out_num_consumed = num_bytes_total;
        return value;
    }